Out-Of-Distribution Detection Is Not All You Need

TL;DR

This paper critiques the reliance on out-of-distribution detection for runtime safety monitoring in neural networks, proposing a focus on detecting incorrect predictions instead, supported by extensive experiments and training data improvements.

Contribution

It introduces the concept of out-of-model-scope detection as a better framework than OOD for runtime monitors and demonstrates its advantages through experiments.

Findings

OOD detection can give false safety impressions

OOD comparison does not identify best error detectors

Removing erroneous data improves monitor training

Abstract

The usage of deep neural networks in safety-critical systems is limited by our ability to guarantee their correct behavior. Runtime monitors are components aiming to identify unsafe predictions and discard them before they can lead to catastrophic consequences. Several recent works on runtime monitoring have focused on out-of-distribution (OOD) detection, i.e., identifying inputs that are different from the training data. In this work, we argue that OOD detection is not a well-suited framework to design efficient runtime monitors and that it is more relevant to evaluate monitors based on their ability to discard incorrect predictions. We call this setting out-ofmodel-scope detection and discuss the conceptual differences with OOD. We also conduct extensive experiments on popular datasets from the literature to show that studying monitors in the OOD setting can be misleading: 1. very good OOD results can give a false impression of safety, 2. comparison under the OOD setting does not allow identifying the best monitor to detect errors. Finally, we also show that removing erroneous training data samples helps to train better monitors.